The advent of DWDM fibre optics telecommunications systems in the early 1990s have enabled a dramatic increase in the transmission capacity over point-to-point links. This was achieved through multiplexing of a large number of individually modulated light beams of different wavelengths onto the same optical fibre. Typical systems installed today would have 64 or more independent channels precisely aligned onto an ITU-T standardized grid at 100 GHz, 50 GHz or even narrower channel spacing. With routine modulation speeds of 10 Gb/s and attaining 40 Gb/s in laboratory experiments, it is not unusual to obtain aggregated capacities in the order of several terabits per second of information being transmitted onto a single optical fibre (S. Bigo, Optical Fibre Communications conference, WX 3, pp. 362–364, Anaheim, 2002). At the same time, electrical switching capacities have been growing at a much slower rate, with current largest electrical switch matrices limited to typically 640 Gb/s in a single stage. Furthermore, the cost of converting the signal from optical to electrical for switching and then back from electrical to optical becomes prohibitively expensive as the number of optical channel increases. All optical switching technologies are therefore becoming more and more attractive to manage the enormous bandwidth being transmitted over optical fibres.
A typical all-optical switch would consist of a large core all-optical switch matrix surrounded by DWDM demultiplexers and multiplexers at each fibre interface. However, for large number of wavelengths channels per optical fibre, this leads to a very large switching core size: for example, a 50 GHz channel spacing system with 128 channels per fibre would require a 1024×1024 switching matrix to switch traffic between 8 incoming fibres to 8 outgoing fibres on a per wavelength basis. Large optical switching matrices are hard to fabricate, complex to control, require overwhelming fibre management and are very expensive. Furthermore, in the absence of wavelength conversion, only a sub-set of the switching matrix capacity is actually in use: each wavelength being switched independently, only 128 8×8 independent connections are used in the 1024×1024 available (0.8% of the overall switching capacity). This huge inefficiency is the primary reason for considering a wavelength switching architecture in which the DWDM demultiplexing and multiplexing are integrated with the switching function.
An example of a wavelength selective all-optical switch is called a wavelength selective cross-connect WSXC (R. E. Wagner, Journal of Lightwave Technology, Vol. 14, No. 6, June 1996, also U.S. Pat. No. 6,097,859) by Solgaard et al. Such a device generally has N incoming fibres and N outgoing fibres, each fibre being capable of transporting M wavelength channels. The WXC enables independent switching of each of the M wavelength channels from the N incoming fibres to the N outgoing fibres. It is functionally equivalent to an input array of N wavelength demultiplexers routed to an output array of N wavelength multiplexers through an array of M N×N optical switches. In such a WXC, there are M×N×N possible optical paths, which is exactly the required flexibility in the absence of wavelength conversion. For example, in the case mentioned above of a 128 channel system at 50 GHz spacing with 8 fibres in and 8 fibres out, the standard large optical core based switch would have over a million possible connections, whereas only 8192 are needed, which is exactly what the WXC architecture enables (128×8×8).